
Leopold Aschenbrenner says "No More Stocks!"
What this covers
This 27-minute Limitless episode has two hosts working through Leopold Aschenbrenner's reported portfolio. The fund is said to be worth over $20 billion, up from a $200 million raise 1.5 to 2 years ago, with no professional investing experience before it started. The hosts begin with the headline positions: a $9 billion short held through puts and options rather than cash shorts, against Nvidia, ASML and Oracle. They set that against a large private stake in Anthropic. They read the short as a rotation, since the picks and shovels semiconductor trade has become crowded. A good part of the discussion goes to Nvidia's $25 billion bond offering, which drew about $85 billion in demand at near-government yields despite $13.7 billion in cash. It landed the same day as an $80 billion buyback and a dividend increase from $0.01 to $0.25.
The middle of the episode covers where money might go next. Aschenbrenner is described as still bullish on specific bottleneck constraints: power, memory, optical networking, and neo-cloud providers such as CoreWeave and Iron, with Core Scientific as a levered way into CoreWeave. The hosts argue that Corning-style optical fiber replaces copper as GPU clusters outgrow it, and that raw materials like copper are a possible later frontier. They also treat power as the most hedged play, and mention SpaceX's orbital data centers as an unconventional answer to permitting limits.
The more contentious turns come late. One host notes Anthropic's reported move from a $60 billion to a $965 billion valuation in about a year. Another raises Jensen Huang's $1.5 billion investment in Marvell and the stock's 70% jump after his Computex remarks, and asks whether that amounts to influence over the market. The closing section asks whether this is a bubble. The hosts say no, arguing that real customers are paying for compute and that human capital and permitting limit the build-out, not financial leverage. They also say copy-trading Aschenbrenner may be worthwhile.
Leopold Aschenbrenner's $9 billion short position in Nvidia and other semiconductor companies signals a rotation in AI investment from overcrowded 'picks and shovels' infrastructure plays toward scarcer bottlenecks (power, memory, optical networking) and direct AI company ownership (Anthropic), not a bubble collapse.
- Nvidia's debt raise despite $13.7B cash reserves and simultaneous $80B buyback authorization indicates cheap capital availability, not distress
- Leopold's long positions in power, memory, and neo-cloud providers (CoreWeave, Iron) alongside his Anthropic stake (20% of fund, now worth ~$7B) reveal where capital is rotating, not a bearish thesis on AI infrastructure itself
- Physical constraints (permitting, human capital, chip fab capacity) prevent the leverage and speculation seen in 2008/dot-com, making this cycle fundamentally different
This asset isn't compiled yet
You're seeing its claims, ranked. Compile it to build the argument threads, weight them, and check each claim against your library — the full view.
Optical fiber providers (Corning and Light) represent the next evolution in data center networking because copper wires become inefficient and generate excessive heat when used to connect very large numbers of GPUs, making optical fiber the necessary next-generation solution.
“typically if you need semiconductor GPU chips to speak to each other, requires like a lot of copper wires. And at some point, when you have so many of these GPUs, copper wires get overheated. There's a lot of heat resistance. You lose a lot of energy. It becomes extremely inefficient”
Assuming Leopold can learn to use the sell button, hedge risk effectively (beyond options-based shorts), and sustain this growth, sustaining performance over 30+ years will be harder than achieving a 30x return.
“dude's been around for 30 years and it turns out being around for 30 years is actually harder than making like a 30X. If you can actually sustain this growth, if you can learn how to sell this push the sell button, learn how to kind of hedge against risk”
On the same day (May 18th), Nvidia authorized an additional $80 billion in share buybacks and increased its dividend from $0.01 to $0.25 per share (25x increase), indicating confidence and capital abundance, not financial distress.
“on May 18th, the board of Nvidia actually authorized an additional $80 billion in buybacks and raised their dividend from a penny to 25 cents per share... multiplied the dividend by 25 times”
Leopold has demonstrated exceptional pattern recognition and predictive accuracy across 1.5 years despite having zero professional investing experience prior to starting his fund, suggesting innate analytical talent.
“He has zero and I must emphasize zero investing experience, right? Um but he's made some of the most amazing calls ever”
Nvidia has the exclusive ability to provide the only 'main machine provider' capable of supporting the 'exponential surplus of demand for AI products,' meaning no other company can effectively replace Nvidia's market position in the near term
“there is a surplus, an exponential surplus of demand for AI products, and there's only one main machine provider that can support that type of demand, and that is Nvidia. So, I don't think they're going anywhere”
Nvidia raised $25 billion in a bond offering (oversubscribed 3-4x with $85 billion in demand) at exceptionally low rates (nearly government bond yields), despite holding $13.7 billion in cash on balance sheet.
“Nvidia just raised $25 billion in a bond offering from external capital”
Copper is the single most critical material for high-bandwidth data transfer over short distances in current data center architecture, and the copper futures market has shown strong performance over the past year
“the copper thing was interesting for me because I learned how critical copper has been recently for a lot of the close data transfer. Copper is the only material that anybody wants. And the only time you stop using copper is when it becomes unviable. When you have to transfer data over too long of a distance, or like you mentioned, it gets too hot. That's when you switch over to fiber...Copper, if you go to that one-year chart, it's probably looking pretty strong”
CoreWeave and Iron are the 'top neo cloud providers' that function similarly to Amazon Web Services (AWS) but specifically for AI companies, offering GPU infrastructure, networking setup, and compute provisioning services
“Coreweave and Iron are basically some of the top neo cloud providers. So, what I mean by that is think of Amazon Web Services, like they create cloud services for any internet company that wants to create a thing. That's what these companies do for any AI company. They have the GPU infrastructure, they set it up for you, they do all the networking such that you don't need to focus on it. You can just train your models, get access to compute, and don't worry about any of the other kind of stuff”
Leopold Aschenbrenner's fund is now worth over $20 billion, with a notional position that has doubled approximately every quarter since initial $200 million raise 1.5-2 years ago.
“his notional position in this fund is now over $20 billion”
Google raised external debt through a bond offering approximately 3 weeks prior to the Nvidia raise and also sold equity just 3 weeks before, suggesting a coordinated pattern among hyperscalers to raise external capital.
“Google did something very similar about uh three weeks ago and then they did it earlier on this year in February... Nvidia, Amazon, Google, and I think three of the other hyperscalers have all raised external debt financing in the last month and a half”
SpaceX is placing data centers in outer space and deploying AI satellites powered by solar energy harvesting, representing an unconventional approach to solving the data center permitting and power constraint problem.
“SpaceX is trying to put them [data centers] into outer space”
Anthropic was valued at $60 billion in March 2025 and is now valued at $965 billion within approximately one year, representing either a 15-16x valuation multiple increase or reflecting the rapid pace of AI company valuation growth
“The first time he invested either privately or through the fund in a VC investment in Anthropic was in March 2025. That was when Anthropic was worth $60 billion. Double digits. Now, Anthropic's recent valuation, they're valued at $965 billion. That is a 15x markup”
Energy and power companies (like Bloom Energy) represent the most hedged AI infrastructure play because electricity demand will grow regardless of whether AI growth accelerates or slows, making them less subject to sector-specific risk.
“energy stocks... no matter what, even in the case that AI demand slows down, energy is still very much in high demand... even in the absence of AI, we need more energy... companies like Bloom Energy deliver on that... the singular trend that is not going to stop no matter what is our demand for energy, electricity, and power”
Leopold's portfolio thesis—outlined in a 65-page 'Situational Awareness' memo ~1.5 years ago when starting his fund—predicted money would rotate from semiconductors and infrastructure specifically into other bottleneck constraints, and this thesis is now actively playing out in real-time.
“he gave us the entire thesis in a 65-page uh AIS A uh about a year and a half ago when he started his fund Situational Awareness, called Situational Awareness, that broke down his entire trade. He even described how the money will rotate from semiconductors and infrastructure specifically into other bottleneck constraints and that's the trade that's playing out today”
Unlike 2008 financial crisis or dot-com bubble, the current AI infrastructure cycle has genuine demand (real customers actually paying for compute), preventing leveraged speculation, and is physically constrained by human capital and permitting, not financial leverage.
“there are people paying for the products that these companies are creating. So, there are real customers buying the products. Back in the dot-com boom, back in the financial crisis, you don't really have this”
When permitting and regulatory approval delays become the binding constraint on data center construction, the companies that can navigate permitting and build infrastructure fastest will capture the most value.
“If anything, just to get the permits to go build it... There are laws, regulations, and a ton of red tape that is currently blocking you”
Leopold has a $9 billion short position (through puts and options, not direct cash shorts) in Nvidia, ASML, and Oracle, representing major infrastructure companies central to the AI buildout thesis.
“we have a total of $9 billion worth of puts in Nvidia, ASML, and Oracle”
The picks-and-shovels AI trade has become extremely overcrowded, with many investors positioned in the same semiconductor and infrastructure suppliers, suggesting money will rotate to less obvious positions.
“the picks and shovel AI trade is extremely overcrowded. And that's what Leopold's most recent portfolio positioning tells us”
The material play at the base of all layers (copper, lithium, etc.) will become increasingly important as infrastructure scales, and yet Leopold has not yet positioned heavily in raw materials, suggesting this is a frontier for the next rotation.
“The materials play is always interesting. That is the the base of all base layers is like, what is the core material that goes in in order to get intelligence out? Copper's one of them, of course lithium. There's a whole bunch of them... Leopold hasn't made his way down there yet and maybe we could front-run him for the next rotational thesis”
Leopold is no longer bullish on the broader semiconductor landscape but remains bullish on specific bottleneck constraints: power, memory, neural cloud providers, and optical networking technologies.
“a month ago, it was revealed that he wasn't especially bullish on the semiconductor landscape. He was still bullish on things like memory and power, those kinds of bottleneck constraints. And maybe he was bullish on neural clouds, as well”
When major influential investors or CEOs publicly endorse or invest in specific stocks (examples: Jensen Huang with Intel, Qualcomm, Marvell, and Weebl, or Donald Trump with various companies), those stocks have historically gone 'up only,' suggesting strong market response to prominent investor/CEO signals
“whenever the people like Jensen, when people like Trump, they say go buy this stock, you probably go buy it. It's paid off in pretty big ways. Like Intel is up how many multiples from the time he bought it? Marvell is up how many multiples from Jensen? It's like these guys know what they're talking about, or at least they have the ability to influence the outcomes of these companies”
Core Scientific is a company that 'helps unlock a lot of Coreweave's infrastructure provision,' and Leopold has privately invested in it, effectively taking a 'levered bet' on Coreweave's infrastructure business growth
“he's privately invested in a company called Core Scientific, which helps unlock a lot of Coreweave's infrastructure provision. So, which means that he's effectively taking a levered bet on Coreweave”
Jensen Huang (Nvidia CEO) made a $1.5 billion investment in Marvell and subsequently stated on stage at Computex that Marvell is 'the next trillion-dollar company,' after which Marvell stock rose 70%, raising questions about insider influence or market manipulation.
“Jensen Huang went on stage at Computex... and he said this is the next trillion-dollar company. Three months prior to him making that statement, Nvidia had made a one and a half billion-dollar investment in Marvell. I don't know where the lines are in terms of like insider trading or market manipulation at this point, but the stock did pump 70% subsequently after he made that statement”
The data and timing of Leopold's trade recommendations make him worth 'copy trading,' meaning his publicly disclosed positions serve as reliable signals for other investors to follow
“I am going to continue copy trading the boy Leopold because dude has just been correct. He's not wrong”
Nvidia's 2019 stock price and recent valuation suggest it has appreciated substantially, but remains a mature, highly-valued company that is being used for debt issuance and capital distribution rather than for aggressive expansion.
“Nvidia stock, it hasn't been doing great. The one you were just showing on screen. It's not like Nvidia's been crushing recently, but it's still like $5 billion, most valuable company in the world. I mean, only down 7% a month is is noteworthy”
The 'next trillion-dollar company' (per Jensen Huang's Computex statement about Marvell) will be a company solving critical infrastructure bottlenecks for AI, not a semiconductor supplier but an infrastructure enabler.
“Jensen Huang... said this is the next trillion-dollar company”
Tesla is developing humanoid robots (potentially for addressing the labor constraint in data center buildout), representing another frontier solution to the human capital bottleneck.
“Who's solving the human problem? Well, Tesla's trying to build humanoid robots, but both of those things are pretty long way out”
SpaceX traded at $230 per share after-hours following the Corsair acquisition, valuing the company at $3.1 trillion, and Elon Musk made more money in one day than Warren Buffett earned in his entire career.
“SpaceX... after hours last night traded at $230 a share. That's 3.1 trillion dollars in value... Elon made more money in a day than Warren Buffett did in his whole career”
Iron stock has appreciated 20% since Leopold's last 13F filing, demonstrating that his neo-cloud infrastructure positions are continuing to perform well
“Iron is one of his biggest positions, and that since his 13F filing is up 20%”
Nvidia's Nvidia stock has declined approximately 7% in the recent period, which is noteworthy in a context where 'everything else has been pumping,' suggesting relative underperformance
“It's not like Nvidia's been crushing recently, but it's still like $5 billion, most valuable company in the world. I mean, only down 7% a month is is noteworthy considering everything else has been pumping”
The speakers believe Leopold's positions are 'probably' likely to continue appreciating and represent a prudent long-term investment strategy given the identified infrastructure trends, though they maintain appropriate uncertainty about continued performance
“I hope it continues. I hope this keeps going. It seems like it's probable to keep going. I think we're all still long. We're all still pretty optimistic and excited, and we'll just continue to evaluate day-by-day”
Leopold's portfolio will 'eclipse the best traders in the world' and 'the best investment funds in the world' if his full AUM growth continues at recent rates and if Nvidia's external debt raises become a sustained trend.
“if Nvidia continues to raise external debt, sell equity potentially in the future if this trend continues, Leopold might be right again. And his his full 40AM will eclipse the best traders in the world the best investment funds in the world”